CBMM, NSF STC » Finding Friend and Foe in Multi-Agent Games [video]
Video
CBMM videos marked with a have an interactive transcript feature enabled, which appears below the video when playing. Viewers can search for keywords in the video or click on any word in the transcript to jump to that point in the video. When searching, a dark bar with white vertical lines appears below the video frame. Each white line is an occurrence of the searched term and can be clicked on to jump to that spot in the video.
Co-authors Max Kleiman-Weiner and Jack Serrino discuss their latest publication where they created a new algorithm. The algorithm that the team developed, dubbed DeepRole, has three components. First, it plays against itself, iteratively, with variation, boosting the likelihood of actions that are beneficial to each player, improving slowly. This has worked well in other games, such as Go and poker. Second, it implements deductive reasoning, based on observed actions, and tracks the belief about the assignment of roles (spy or non-spy) given the history of actions. This inference, a computational mechanism to distinguish friend and foe from actions, is a type of Theory of Mind. It was Serrino who, as an Avalon enthusiast, brought detailed insight into how people actually play the game, which helped build the inferential computation. The space of Avalon actions is too large for brute-force search, so in the third element of the algorithm (as in successful poker AI) a deep neural network is used to prune the search tree.
have an interactive transcript feature enabled, which appears below the video when playing. Viewers can search for keywords in the video or click on any word in the transcript to jump to that point in the video. When searching, a dark bar with white vertical lines appears below the video frame. Each white line is an occurrence of the searched term and can be clicked on to jump to that spot in the video.
